[2016-Sep-30] Working with reports based on historical data is like driving a car while looking in the rear view mirror. I heard this term again when I went to the Microsoft Data Science Summit in Atlanta (https://ignite.microsoft.com/microsoftdatascience) this past week. Which gave me a second thought to consider: how do we define a historical fact and what sets a current time frame from anything else (i.e. history).
Then I read good article "Myth: Managing by historical data is like driving while looking in the rear view mirror" by Allan Wille and realized that I'm not alone in this camp of doubters in the real meaning of the ineffective rear view mirror driving vs. better front driving for business metrics analytics.
In this article Allan outlined three types of a business performance analytics:
1) Current performance (real-time or near-real-time)
2) Timeframe-appropriate historical performance data
3) Predictive or conditional alerts (based on goals, history, or environmental factors).
And I would go even further, I believe all reporting that we currently poses with our modern existent technology is based on historical data. Stream analytics is based on some near-real-time events but they still happen a few moments ago, they're already history. Predictive analytics which aims to predict future events are based on analytical models which usually are trained and tested using historical data sets. Even in cognitive analysis we will have to use a large data repository that vendors may have collected in order to work with image or speech recognition technologies.
My middle school teacher of physics had planted this seed of doubt a while ago when she confronted our class with an idea that a car speedometer never showed a real-time vehicle speed. Because there is always a latency between different components of this mechanism. And speed measurement, that I can observe at the moment, may simply be a real representation of the speed fact a few moments ago.
So, if I want to become a data scientist, I will start defining what I mean by a historical fact, at least to myself. Otherwise I may assume a car in my rear-view mirror as history while it overtakes me and starts showing its rear lights in my front view instead.
Then I read good article "Myth: Managing by historical data is like driving while looking in the rear view mirror" by Allan Wille and realized that I'm not alone in this camp of doubters in the real meaning of the ineffective rear view mirror driving vs. better front driving for business metrics analytics.
In this article Allan outlined three types of a business performance analytics:
1) Current performance (real-time or near-real-time)
2) Timeframe-appropriate historical performance data
3) Predictive or conditional alerts (based on goals, history, or environmental factors).
And I would go even further, I believe all reporting that we currently poses with our modern existent technology is based on historical data. Stream analytics is based on some near-real-time events but they still happen a few moments ago, they're already history. Predictive analytics which aims to predict future events are based on analytical models which usually are trained and tested using historical data sets. Even in cognitive analysis we will have to use a large data repository that vendors may have collected in order to work with image or speech recognition technologies.
My middle school teacher of physics had planted this seed of doubt a while ago when she confronted our class with an idea that a car speedometer never showed a real-time vehicle speed. Because there is always a latency between different components of this mechanism. And speed measurement, that I can observe at the moment, may simply be a real representation of the speed fact a few moments ago.
So, if I want to become a data scientist, I will start defining what I mean by a historical fact, at least to myself. Otherwise I may assume a car in my rear-view mirror as history while it overtakes me and starts showing its rear lights in my front view instead.
